基于状态加权MTF与多尺度轻量化网络的齿轮箱故障诊断

Gearbox Fault Diagnosis Based on State-Weighted MTF and Multi-Scale Lightweight Network

摘要:

针对齿轮箱微弱故障提取困难及深层诊断模型计算复杂度高的问题，对基于状态加权马尔可夫转移场与多尺度轻量化网络的诊断方法进行了研究；采用了状态加权策略的关键技术和方法，对马尔可夫转移概率进行了重标定计算，以增强隐蔽故障特征的区分度；其技术创新和独特之处在于构建双分支GhostNet轻量化网络，引入多尺度特征增强与通道注意力机制，利用不同尺寸感受野协同提取并融合故障特征；经实验测试实现了在精度、轻量化及抗噪性能上显著优于多种对比模型的结果，验证了变工况下的泛化能力；经实际应用满足了降低计算开销并实现高精度诊断的要求，为边缘设备实时监测提供了可行的工程上的应用。

Abstract:

Aiming at the difficulties in extracting weak fault features of gearboxes and the high computational complexity of deep models, research was conducted on a diagnosis method based on state-weighted Markov transition field and multi-scale lightweight network; the key technologies and methods of state weighting strategy were adopted, and recalibration calculation was conducted on transition probabilities to enhance the distinction of hidden fault features; the technical innovation and uniqueness lie in constructing a dual-branch GhostNet lightweight network and introducing multi-scale feature enhancement and channel attention mechanisms to collaboratively extract and fuse fault features; superior results in accuracy, lightweightness, and anti-noise performance were achieved through experimental testing, verifying the generalization ability under variable working conditions; the requirements of reducing computational overhead and achieving high-precision diagnosis were satisfied through practical application, providing a viable engineering application for real-time monitoring of edge devices.